Genetic variation is abundant across individuals within a population, and it can explain a substantial proportion of the risk of disease and variation in quantitative phenotypes. An interesting thing about genetic variation is that it can be treated like a natural experiment, in which observing the genotype at a relevant locus is akin to observing the outcome of a randomly allocated treatment effect. This is the basis of Mendelian randomisation, a method for inferring causal relationships between phenotypes. Much of my current work focuses on developing statistical and computational methods to improve the scope and reliability of Mendelian randomisation.